IA-ENI/NeuralNetwork/NeuralNetworkPCL/NeuralSystem.cs

using System;

namespace NeuralNetworkPCL
{
    public class NeuralSystem
    {
        DataCollection data;
        NeuralNetwork network;
        IHM ihm;

        // Configuration
        double learningRate = 0.3;
        double maxError = 0.005;
        int maxIterations = 10001;

        public NeuralSystem(int _nbInputs, int _nbHidden, int _nbOutputs, String[] _data, double _trainingRatio, IHM _ihm)
        {
            data = new DataCollection(_data, _nbOutputs, _trainingRatio);
            network = new NeuralNetwork(_nbInputs, _nbHidden, _nbOutputs);
            ihm = _ihm;
        }

        public void LearningRate(double _rate)
        {
            learningRate = _rate;
        }

        public void MaximumError(double _error)
        {
            maxError = _error;
        }

        public void MaximumIterations(int _iterations)
        {
            maxIterations = _iterations;
        }

        public void Run()
        {
            int i = 0;
            double totalError = Double.PositiveInfinity;
            double oldError = Double.PositiveInfinity;
            double totalGeneralisationError = Double.PositiveInfinity;
            double oldGeneralisationError = Double.PositiveInfinity;
            Boolean betterGeneralisation = true;

            while (i < maxIterations && totalError > maxError && betterGeneralisation)
            {
                oldError = totalError;
                totalError = 0;
                oldGeneralisationError = totalGeneralisationError;
                totalGeneralisationError = 0;
                
                // Evaluation
                foreach (DataPoint point in data.Points())
                {
                    double[] outputs = network.Evaluate(point);
                    for (int outNb = 0; outNb < outputs.Length; outNb++)
                    {
                        double error = point.Outputs[outNb] - outputs[outNb];
                        totalError += (error * error);
                    }

                    // Calcul des nouveaux poids par rétropropagation
                    network.AdjustWeights(point, learningRate);
                }

                // Généralisation
                foreach (DataPoint point in data.GeneralisationPoints())
                {
                    double[] outputs = network.Evaluate(point);
                    for (int outNb = 0; outNb < outputs.Length; outNb++)
                    {
                        double error = point.Outputs[outNb] - outputs[outNb];
                        totalGeneralisationError += (error * error);
                    }
                }
                if (totalGeneralisationError > oldGeneralisationError)
                {
                    betterGeneralisation = false;
                }

                // Changer le taux
                if (totalError >= oldError)
                {
                    learningRate = learningRate / 2.0;
                }

                // Information et incrément
                ihm.PrintMsg("Iteration n°" + i + " - Total error : " + totalError + " - Gener Error : " + totalGeneralisationError + " - Rate : " + learningRate + " - Mean : " + String.Format("{0:0.00}", Math.Sqrt(totalError/data.Points().Length),"%2"));
                i++;
            }
        }
    }
}